From 56b4ea7e7a8fe29c1c1066def42c792cb51540d0 Mon Sep 17 00:00:00 2001
From: Guus Sliepen <guus@meshlink.io>
Date: Tue, 9 Feb 2021 21:00:35 +0100
Subject: [PATCH] Try to recover from select() returning an error.

If we get an error back from select(), it could be because one of the
filedescriptors is in a bad state. This shouldn't occur, but if it does,
just call all the registered I/O callbacks to have them check for errors.
Only quit the event loop if the error persists.

Also fix some I/O callbacks that didn't expect to be called when the
filedescriptor isn't signalled.
---
 src/event.c      | 60 +++++++++++++++++++++++++++++++++++++++++++++++-
 src/net_setup.c  | 27 ++++++++++++++++++----
 src/net_socket.c |  4 ++++
 3 files changed, 86 insertions(+), 5 deletions(-)

diff --git a/src/event.c b/src/event.c
index a7c1f25a..82133eb4 100644
--- a/src/event.c
+++ b/src/event.c
@@ -211,6 +211,10 @@ static void pipe_init(event_loop_t *loop) {
 	assert(result == 0);
 
 	if(result == 0) {
+#ifdef O_NONBLOCK
+		fcntl(loop->pipefd[0], F_SETFL, O_NONBLOCK);
+		fcntl(loop->pipefd[1], F_SETFL, O_NONBLOCK);
+#endif
 		io_add(loop, &loop->signalio, signalio_handler, NULL, loop->pipefd[0], IO_READ);
 	}
 }
@@ -278,11 +282,56 @@ void idle_set(event_loop_t *loop, idle_cb_t cb, void *data) {
 	loop->idle_data = data;
 }
 
+static void check_bad_fds(event_loop_t *loop) {
+	// Just call all registered callbacks and have them check their fds
+
+	do {
+		loop->deletion = false;
+
+		for splay_each(io_t, io, &loop->ios) {
+			if(io->flags & IO_WRITE) {
+				io->cb(loop, io->data, IO_WRITE);
+			}
+
+			if(loop->deletion) {
+				break;
+			}
+
+			if(io->flags & IO_READ) {
+				io->cb(loop, io->data, IO_READ);
+			}
+
+			if(loop->deletion) {
+				break;
+			}
+		}
+	} while(loop->deletion);
+
+	// Rebuild the fdsets
+
+	memset(&loop->readfds, 0, sizeof(loop->readfds));
+	memset(&loop->writefds, 0, sizeof(loop->writefds));
+
+	for splay_each(io_t, io, &loop->ios) {
+		if(io->flags & IO_READ) {
+			FD_SET(io->fd, &loop->readfds);
+			io->cb(loop, io->data, IO_READ);
+		}
+
+		if(io->flags & IO_WRITE) {
+			FD_SET(io->fd, &loop->writefds);
+			io->cb(loop, io->data, IO_WRITE);
+		}
+
+	}
+}
+
 bool event_loop_run(event_loop_t *loop, pthread_mutex_t *mutex) {
 	assert(mutex);
 
 	fd_set readable;
 	fd_set writable;
+	int errors = 0;
 
 	while(loop->running) {
 		clock_gettime(EVENT_CLOCK, &loop->now);
@@ -338,10 +387,19 @@ bool event_loop_run(event_loop_t *loop, pthread_mutex_t *mutex) {
 			if(sockwouldblock(errno)) {
 				continue;
 			} else {
-				return false;
+				errors++;
+
+				if(errors > 10) {
+					return false;
+				}
+
+				check_bad_fds(loop);
+				continue;
 			}
 		}
 
+		errors = 0;
+
 		if(!n) {
 			continue;
 		}
diff --git a/src/net_setup.c b/src/net_setup.c
index 269b46ed..6168c4ca 100644
--- a/src/net_setup.c
+++ b/src/net_setup.c
@@ -332,6 +332,25 @@ int setup_tcp_listen_socket(meshlink_handle_t *mesh, const struct addrinfo *aip)
 	fcntl(nfd, F_SETFD, FD_CLOEXEC);
 #endif
 
+#ifdef O_NONBLOCK
+	int flags = fcntl(nfd, F_GETFL);
+
+	if(fcntl(nfd, F_SETFL, flags | O_NONBLOCK) < 0) {
+		closesocket(nfd);
+		logger(mesh, MESHLINK_ERROR, "System call `%s' failed: %s", "fcntl", strerror(errno));
+		return -1;
+	}
+
+#elif defined(WIN32)
+	unsigned long arg = 1;
+
+	if(ioctlsocket(nfd, FIONBIO, &arg) != 0) {
+		closesocket(nfd);
+		logger(mesh, MESHLINK_ERROR, "Call to `%s' failed: %s", "ioctlsocket", sockstrerror(sockerrno));
+		return -1;
+	}
+
+#endif
 	int option = 1;
 	setsockopt(nfd, SOL_SOCKET, SO_REUSEADDR, (void *)&option, sizeof(option));
 
@@ -528,8 +547,8 @@ static bool add_listen_sockets(meshlink_handle_t *mesh) {
 		for(int i = 0; i < mesh->listen_sockets; i++) {
 			io_del(&mesh->loop, &mesh->listen_socket[i].tcp);
 			io_del(&mesh->loop, &mesh->listen_socket[i].udp);
-			close(mesh->listen_socket[i].tcp.fd);
-			close(mesh->listen_socket[i].udp.fd);
+			closesocket(mesh->listen_socket[i].tcp.fd);
+			closesocket(mesh->listen_socket[i].udp.fd);
 		}
 
 		mesh->listen_sockets = 0;
@@ -617,8 +636,8 @@ void close_network_connections(meshlink_handle_t *mesh) {
 	for(int i = 0; i < mesh->listen_sockets; i++) {
 		io_del(&mesh->loop, &mesh->listen_socket[i].tcp);
 		io_del(&mesh->loop, &mesh->listen_socket[i].udp);
-		close(mesh->listen_socket[i].tcp.fd);
-		close(mesh->listen_socket[i].udp.fd);
+		closesocket(mesh->listen_socket[i].tcp.fd);
+		closesocket(mesh->listen_socket[i].udp.fd);
 	}
 
 	exit_requests(mesh);
diff --git a/src/net_socket.c b/src/net_socket.c
index abf6a7cf..22cbe26f 100644
--- a/src/net_socket.c
+++ b/src/net_socket.c
@@ -524,6 +524,10 @@ void handle_new_meta_connection(event_loop_t *loop, void *data, int flags) {
 	fd = accept(l->tcp.fd, &sa.sa, &len);
 
 	if(fd < 0) {
+		if(sockwouldblock(errno)) {
+			return;
+		}
+
 		if(errno == EINVAL) { // TODO: check if Windows agrees
 			event_loop_stop(loop);
 			return;
-- 
2.39.5